Advanced topic 9, 2012
User-Centric Management of Wireless LANs The content of this wiki page was adapted from the paper above. All credit goes to original authors. Topic 9 from Advanced Topics 2012 Introduction In the past few years, the deployment of Wireless Local Area Networks (WLANs) has been hugely enhanced, which in turn, the density deployment of WLAN users and access points (APs)3 are increasing at a staggering rate. As a result, a user finds an increasing number of APs deployed within its sensing range, but the user is allowed to select only one AP to be an association before joining in a WLAN. However, providing the effective network management to optimize users’ throughput all the time is becoming an urgent issue of high-density WLANs due to the factor that after the association, the throughput user achieved depends on the load delivered by the selected AP and the neighboring channel conditions567. Traditionally, the network management framework of WLANs is AP-centric. It has following characters. *Exclusively AP one-side control the transmission and none much related to users *The procedures are simple and workable *Helpless to enhance the throughput *Not suitable to support different QoS in some cases But nowadays, User-centric network management framework is more frequently used1 *The WLAN management is sharing pipeline between users and APs *Maximize users’ throughput *Optimize AP aggregate throughput and dynamic APs’ channel selection *Network conditions and access priority need to be considered *User actively senses the network conditions and delivers the sensed results with its access priority to the AP candidates Overview of Network Management in WLANs A. Basic management entities and activities There are three main management entities defined in IEEE 802.11 2family standards: *Entities: *Station management (SME) *MAC Sub-layer Management Entity (MLME) *Physical Layer Management Entity (PLME) The dominant functions of MLME and PLME are to in-phase, managing power, combination and recombination, which is usually used to process latency in network device firmware, being more suitable to be executed in hardware instead of network management. SEM is to gather the inter-layers status and to configure specific parameters. It is the central entity in network management. Meanwhile, MLME and PLME provide some system status to SME by interfaces entity in network management. Meanwhile, MLME and PLME provide some system status to SME by interfaces between MLME and SME. There are also three activities of APs: *Activities: *Parameter configuration *Users information management *Network monitoring Specifically, parameter configurations include channel selection and power configuration; user information management contains AP association lists and management; and network monitoring relates to both downlink and uplink monitoring traffic from users to APs. Moreover, there are two types among those activities. Passive activities include the priority management and the network monitoring while positive activities include the parameter configuration and the AP association lists. Additionally, during network deployment, power configuration can be set without much change to it. Therefore, the scope of AP association and channel selection is critical when it comes to most parts of network management. B. AP Association and Channel Selection Firstly, Fig. 1 shows the basic working flow over the network management framework in traditional WLANs. As seen from the Fig.1, an AP needs to select a channel either manually4 or via the configuration of neighboring APs within its sensing coverage. Then, the configuration information will be delivered in the beacon frames to those users nearby. However, the user is more likely to only select the AP from those access points with enough signal power even if this user can sense and know the signal power in all the transmissions within the sensing coverage. At last the AP adds the user into its serving list based on whether the user is admitted with secure credentials like the permissible MAC address . Furthermore, the AP association and the channel selection do not use AP-related or user-related information, so they tend to be AP-centric. User-centric network management framework The fundamental idea of user-centric network management framework, instead of traditional AP-centric network management framework, is to involve the users participation in network management framework. This framework can solve the throughput problems AP-association left. The basic principle is to provide the pipeline information exchange platform and duplexing selection between APs and users that enable to enhance traditional APs one-side control to consider what a user needs. The detailed user-centric network management working procedures illustrate below. First of all, AP selects a channel manually or based on neighboring APs’ sensing range. After that, AP delivers configuration information (like channel, data rate, and MAC address) periodically to users by beacon within its sensing range. The user, collecting the APs status within its receiving range periodically, considers the network conditions (like beacon frame transmission and data transmission) which is also as network snapshot and piggybacks the network snapshot, access priority and association request frame to all candidate APs in its receiving range. AP, which receives the frame, calculates the potential throughput for users by users provided information and transmission status by itself and returns the response frame with calculated results to the users by piggyback. Then the user will choose the AP that provides the maximum potential throughput and inform it with a new association request. The selected AP replies with a response frame to finalize the association and stores the users’ network snapshot. It is worthy mentioning that AP not only can provide the potential throughput and be selected by users but also can dynamically select the operational channel and optimize the aggregate throughput by the collected user information and snapshot under user-centric network management system. Moreover, users can select the AP that can provide the maximum throughput, avoiding being passive selected by AP. Theoretical analysis The theoretical analysis is introduced to illustrate why the user-centric network management framework is the superior approach than the traditional network management framework in AP association and channel selection. Because the performance of p-persistent access model 8 is very similar to the IEEE 802.11MAC contention access mechanism, the former is used to simulate the performance of the latter. In this paper, the downlink traffic is the unique considered in the analysis since the traffic on downlink takes the most percentage of the total network traffic. This section is divided into three parts which are the system model, optimal AP association and the optimal channel selection. A. System model Firstly, the system model is illustrated as follows. In order to calculate the throughput some fundamental knowledge about this system is briefly introduced. The packet collision is the only factor considered to affect the failed transmissions. According to CSMA mechanism, the transmission of packet fails if more than one accessed points in the sensing area transmit simultaneously. In other words, only one AP is allowed to transmit in the same sensing range. Hence, the probability of a successful transmission from AP ���� to user ���� in a time slot which is operating on channel ��is given by: (1) Where, ���� denotes the probability of the transmission of ����, and ������ denotes the probability that ���� sends packets to ����. Hence, the probability that ���� senses channel �� busy is obtained by: (2) Therefore, the throughput of ���� linking with ���� is defined as the average payload size transmitted in the unit time and operates on channel ��. The formula of the throughput is given by: (3) where, E�� indicates the average payload size of a data packet, and ��busy and ��idle denote the average length of a busy time slot and an idle time slot, respectively. Substituting (1) and (2) into (3), the formula of the throughput is changed to: (4) Where, there is a relationship between ������ and these parameters which are ����, the number of users linked with ����, the traffic load of each user, and the priority of each user. This relationship is derived as follows. Since AP usually maintains the independent access queue of each user in its sensing range, there are two virtual contention mechanisms utilized to share the physical radio, Distributed Coordination Function (DCF) and Enhanced DCF (EDCF), which is described below. Under the virtual contention mechanism of DCF, a user selects a back-off window 10 from ����min, where ����min denotes the minimum contention window. The packet is transmitted when the back-off window reduces to zero. If the packet transmission fails because of the contention, the user randomly chooses a new back-off window from a double extended range 2����min until the contention window achieves the maximum contention window ����max. If the successful packet transmission is recovered, the contention window returns to ����min immediately. If the EDCF is used as the virtual contention mechanism, the process is different with that with DCF. The different minimum and maximum contention windows have different priorities under the EDCF mechanism. The level of the priority is inversed to the values of the minimum and maximum contention windows. The transmit probability of the queue for ���� with EDCFcan be obtained as: (5) Where, �������� stands for the average contention window for the access queue to ��j 11. ������ indicates the load of the queue for ���� associated with ����. The transmit probabilities depend on different virtual contention mechanism. Firstly, the CWij is equal in all queues owing to the whole users under the IEEE 802.11 DCF sharing the equal priority. Therefore, the transmit probability under this situation is given by: (6) Where, (7) Secondly, if the users under IEEE 802.11e EDCF virtual contention mechanism, the queues with different priorities have different minimum and maximum contention windows, which results that the average contention windows is different. 13 �������� is given by: (8) Where, Substitute (8) into (7), the pi,c can be obtained when ����, �������� min and ��������max are given. Substitute (8) into (5), the value of pij can be obtained. These analytical results are very useful to make a design on the optimal AP association and the optimal channle selection. B. Optimal AP Association In the second section, the optimal AP association is discussed. As the description before, AP association is implemented when a new user joins the network or handoff because the previous AP is unable to achieve the performance requirement. The optimal AP association corresponding to the AP would provide the maximum throughput to a user. The mathematical model of the throughput is given by: (9) substituting (4) into (9): (10) From the formula (10), the throughput may fluctuate with neighbors’ network conditions and the transmission probability ������. Hence, the information pipeline under the user-centric framework becomes significant important on the optimal AP association. The neighbors’ network conditions under the traditional network management framework are usually available. Therefore, the throughput for user ���� is obtained as: (11) However, this throughput level is not the optimal. C. Optimal Channel Selection The third section demonstrates the Optimal Channel Selection. In the traditional network management framework, only the transmit probability pij of the AP is available for channel selection. Hence, the channel selection is based on the network load of neighboring APs, which is proportion to pij. Compared to the traditional network management framework, AP under the user-centric framework is able to collect the network condition by the associated users. The optimal channels could be able to be selected to maximize the aggregate throughput. The aggregate throughput for AP under the user-centric framework is given by: (12) Therefore, it is impossible for the traditional network management framework to achieve the maximum total throughput by channel selection since the situation that the sensing areas of the access point and the user linked to it change with different locations does not always hold. However, the user-centric framework is able to dynamically collect the network conditions to achieve the optimal channel selection and obtain the maximum aggregate throughput. Implementing the Management Framework AP association and channel selection activities will be conducted only when a new user is added into the network or the association hand-off as the throughput performance is out of the requirement. A little change of operating channel or the associated users may lead to the network management activities redo again. Hence, the convergence of the iterative management procedures of these two activities needs to be done before the user-centric network framework is implemented into the real network. In this section, a hysteresis in the implementation of the management procedures is illustrated to address the oscillatory behavior resulted from the optimal AP association and the channel selection. A.The dual-threshold triggering conditions A new association between AP and user triggered by users only when the existing association throughput is lower than a lower-bound threshold or a new (optimal) AP-association potential throughput is much stronger than its current throughput by at least a minimum improvement threshold. Specifically, the user executes a new AP association procedure only if: A new association triggered by AP happens only when the current aggregate downlink throughput is lower than a lower-bound threshold or a new (optimal) channel selection potential throughput is much stronger than its current throughput by at least a minimum improvement threshold. Specifically, AP establishes a new AP association procedure only if: B.The dual period operation AP association or channel selection should be taken place periodically to constantly optimize to such operation. Then a comparatively short T is selected in order to have an initialization period during each Ta procedure, and users who have been connected to the AP need to recalculate the throughput that may be potentials to them. Since the operation of channel selection will lead to the reconnection of all connected users, select a longer Tc (decision period) rather than Ta (shorter period), like the previous procedure, the AP assesses its general performance and determines whether to set a new channel selection. At last, �� ∈ (0, 1) and �� ∈ (0, 1) acted as the smoothing factors, depending on the realistic considerations can be introduced to get the average traffic ups and downs: Where，�� and �� normally are set to 0.9. Additionally, ������, and ������,��+���� stand for updated throughput of ���� at time t and the average throughput in the time range �� + ����) respectively, which can be used to define ����,�� and ����,��+���� as well. The throughput of the APs and the connected users should be updated at each instant decision time and see if the triggering conditions are satisfied, then the finest channel selection described above can be used. However, even though the short period of update intervals can result in more throughput, hard breakdowns caused from each recombining period in transmission will affect users because of the high frequency in recombining. The effect on hard breakdowns has to be assessed through specifying a WLAN APs, in order to analyze the simulated deployment and then get the value for ���� = 600s and ���� = 1800s. Obviously, the lower-bound can be set as follows since the choice of it is in linear proportion to the average throughput: Hence, it can be seen that the calculation depends on the link transmission speed (R) and the number of users (Nu). And CR∈ (0, 1) is a mapping coefficient, while NA is the number of APs. For example, if R = 54 Mbps, then set CR = 0.5. Meanwhile, thresholds are improved aimed at enlarging the throughput and the choice of the enhanced thresholds will be determined by the period time of breakdown. Here, set these thresholds as follows for later simulations: Where Tb stands for the breakdown period and C is a constant which is always larger than 1, determining the improved extension. Simulations The simulation uses the software [http://en.wikipedia.org/wiki/NS2 NS2 13 to analyze the user-centric network management framework throughput performance. The parameters can be seen as follows: All the APs and users are randomly and uniformly distributed with the density of one AP per unit area, meanwhile the user density varies with the simulation scenarios. A : Effect of Access Priority * The figure shows the available throughput of the user with high access priority is always greater than that with low access priority. And the available throughput for the considered user decreases with the number of users with high priority no matter what access priority it has. *The result above illustrates the access priority distribution in the AP affects the potential through, and since the throughput attenuation as the number of users increases, the AP association can play a vital role in increasing the throughput for a user. B : AP Association over User-centric Network Management Framework *The figure shows that AP association based on the user-centric network management framework always outperforms other frameworks, and the throughput gain ranges from 25''.0% to 90.5%14. *The result above illustrates the throughput comes from access priority and the network conditions. A new AP association with less high priority users will provide better throughput if the current AP has a lot of users with high priority. Moreover, the throughput can be increased by associating with a new AP with better network conditions, due to the poor AP network conditions. C: Channel Selection over User-centric Network Management Framework * *The figure shows the user-centric management provides the highest throughput -max 69.3%, While the throughput becomes similar for all the systems as the APs get saturated with load. *The result above illustrates that the channel selection in user-centric management framework is initiated by the AP. If the channel condition is weak, the AP then decides to change the channel to be a potentially better channel. The channel can provide poor throughput for several reasons, like channel selective fadings or channel collisions with neighboring APs. However, as the APs get saturated with users, the throughput performances for all the management systems tend to be the same. D : Impact of Dual-threshold in Implementation *CR∈'' (0, 1) is a mappin g factor for available lower-bound throughput and physical transmitting speed C>1 is a constant for leveraging the improvement. *The figure shows the throughput increases as CR increases, and decreases as C increases. When CR< 0.3, the throughput is very small. *The result above illustrates the management operation will not be triggered by a small lower bound threshold or a large value of a minimum improvement threshold. Conclusion Traditional and Self organization management framework tries to optimize the users throughput while not looking at the whole scenario. They ignore one very important perspective, which is users' perspective. User-centric NM framework utilizes the information from the users to deliver better throughput to the users; it is a win-win situation. And, the simulations backs up this model with significant increase in throughput gain. Although the security issues with this framework is somewhat unattended which leaves open the potential security vulnerability from fake user snapshot, this framework also has the most potential to live upto the demand of high density wireless networks. Reference 1 Y. Zhu, Q. Ma, C. Bisdikian, C. Ying, “A user-centric network management framework for high-density wireless LANs," in Proc. 2009 IFIP/IEEE International Symp. Integrated Netw. Management, pp. 218-225, 2008. 2 IEEE 802.11, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications, IEEE Standard 802.11, 1999. 3 A. Akella, G. Judd, S. Seshan, and P. Steenkiste, “Self-management in chaotic wireless deployments," in ''Proc. 11th Annu. Int. Conf. Mobile Comput. Netw., ''2005, pp. 185-199. 4 Y. Hua, Y. Zhu, and Z. Niu, “A channel assignment scheme in high density WLANs to mitigate pesudo capture effect," in ''Proc. IEEE ''GLOBECOM, ''2007, pp. 4796-4800. 5 L. Wang, Z. Niu, Y. Zhu, H. Deng, and M. Yano, “Integration of SNR, load and time in handoff initiation for wireless LAN," in Proc. IEEE PIMRC, 2003, pp. 2032-2036.  6 G. Kasbekar, J. Kuri, and P. Nuggehalli, “Online association policies in IEEE 802.11 WLANs," in Proc. 4th Int. Symp. Modeling Optimization Mobile, Ad Hoc, Wireless Netw., 2006, pp. 1-10.  7 V. A. Siris and D. Evaggelatou, “Access point selection for improving throughput fairness in wireless LANs," in Proc. 10th IFIP/IEEE Int. Symp. Integrated Netw. Management, 2007, pp. 469-477. 8 L. Kleinrock and F. A. Tobagi, “Carrier sense multiple-access modes and their throughput-delay characteristics," IEEE Trans. Commun., vol. 23, pp. 1400-1416, 1983. 9 IEEE 802.11e, Part 11: Wireless Medium Access Control (MAC) and physical layer (PHY) specifications: edium Access Control (MAC) Enhancements for Quality of Service (QoS), IEEE 802.11e, 2005. 10 F. Cali, M. Conti, and E.Gregori, “Dynamic tuning of the IEEE 802.11 protocol to achieve a theoretical throughput limit," IEEE/ACM Trans. Netw., vol. 8, no. 6, pp. 785-799, Dec. 2000 11 G. Bianchi, “Performance analysis of the IEEE 802.11 distributed coordination function," IEEE J. Sel. Areas Commun., vol. 18, no. 3, pp. 535-547, Mar. 2000. 12 Y. Zhu and Z. Niu, “Dynamic polling mechanism for enhancing voice transmission in IEEE 802.11e wireless LANs," Int. J. Wireless Mobile Comput., vol. 1, no. 3/4, pp. 252-259, 2006. 13 K. Fall and K. Varadhan, “Ns notes and documentation," the VINT Project, UC Berkeley, LBL, USC/ISI, and Xerox PARC, Nov. 1997. 14 B. Kauffmann, F. Baccelli, A. Chaintreau, V. Mhatre, K. Papagiannaki, and C. Diot, “Measurement-based self organization of interfering 802.11 wireless access networks," in Proc. IEEE INFOCOM, 2007, pp. 1451-1459.